Layering a computing infrastructure. Middleware. The new infrastructure: middleware. Spanning layer. Middleware objectives. The new infrastructure

Transcription

1 University of California at Berkeley School of Information Management and Systems Information Systems 206 Distributed Computing Applications and Infrastructure Layering a computing infrastructure Middleware Applications Application components Middleware by David G. Messerschmitt Operating system Network 2 Spanning layer Application Distributed object management Windows NT TCP UDP Mac OS TCP UDP UNIX TCP UDP Internet protocol Network 1 Network 2 Spanning layers 3 The new infrastructure: middleware Layer of software between OS and application Hides heterogeneity Provides generic common services Increases level of abstraction By its nature, not single platform, nor bundled with equipment Microsoft attempts to be an exception 4 The new infrastructure Middleware is where new capabilities are added by layering where much of the experimentation and innovation happens in the infrastructure where the successful approaches have a chance to become a spanning layer and/or integrated into a distributed OS Boundary to a distributed OS is fuzzy 5 Middleware objectives Hide heterogeneity Location independence Common functionality needed by many applications Software portability and mobile code Help integrate legacy facilities Aid application interoperability Aid scalability 6 1

2 Some middleware categories Transaction processing Simplify the coordination of complementary resource managers Message-oriented middleware Support message and queuing capabilities where resource mangers are not available simultaneously (like workflow) Some middleware categories Distributed object management Support applications that are distributed across heterogeneous platforms and organizations Mobile code Allow application code to be moved and executed on heterogeneous platforms Without prior software installation 7 8 Infrastructure software today With networks, new emphasis on: Portability: applications run across multiple platforms (avoid lock-in) Interoperability: pieces of application must work together (benefit from network effects) What are some examples of each? University of California at Berkeley School of Information Management and Systems Information Systems 206 Distributed Computing Applications and Infrastructure Mobile code and Java by David G. Messerschmitt 9 Reminder: two key requirements With networks, new emphasis on: Portability: applications run across multiple platforms (avoid lock-in) Interoperability: pieces of application must work together (benefit from network effects) Java emphasizes portability Dynamic portability: mobile code Mobile code: Code representing a software program that can be moved to heterogeneous platforms and executed there ❶ Send code (as a message) to a host ❷ Execute the program represented by that code

3 Portability can aid interoperability Mobile code originating from a common source can enhance interoperability 13 Java Portability Write once, run anywhere Programming productivity Garbage collection (no memory leaks) Multi-threaded Scalability Move execution cycles Interoperability Software components come from common repository 14 Mobile code: Code representing a software program Idea of mobile code Some mobile code advantages MA ❶ Send as a message to a host Executing program closer to user can enhance interactivity Mobile code originating from a common source can enhance interoperability and bypass network effects Mobile agent: Code and data representing an object or component ❷ Execute the program Shifting location of computation can enhance scalability ❶ Agent launched Idea of mobile agents ❷ Agent executes in each host, modifying its state MA Java virtual machine Java program Mobile code Compilation ❸ Agent returns JIT compiler VM interpreter Bytecode: low level but machine independent Native machine instructions

4 VM as spanning layer SUN/Java strategy Applications JavaBeans component framework Java virtual machine Operating system Java VM spanning layer License Java freely, even to rival Microsoft Why? License terms give Sun a modicum of control over the standard Why? How does Sun expect to make money? University of California at Berkeley School of Information Management and Systems Information Systems 206 Distributed Computing Applications and Infrastructure Distributed object management by David G. Messerschmitt Distributed object management Emphasis is on interoperability Allows objects on one host to invoke methods of objects on another host Platform, language independent CORBA vs DCOM Portability is not the emphasis 22 Interoperability Interoperability Client Client object can invoke methods of server object, even if they are running on different platforms, written in different languages What else is needed? 23 Client Interoperability also requires: Common structure of data Common interpretation of data Agreement on protocols 24 4

5 Before and after What is the acronym? OS 1 Distributed application OS 2 OS 3 IP Distributed application Distributed object management middleware OS 1 OS 2 OS 3 IP 25 Potential spanning layer Spanning layer Common Object Request Broker Architecture or Concerned Off-Road Bicyclist Association? 26 Client objects CORBA architecture objects Object request broker core Object adapter Location-independent application Object Management Group CORBA standard User datagram protocol (UDP) Protocol layer Internet protocol (IP) Application Object request broker Internet Inter-ORB Protocol (IIOP) Transmission control protocol (TCP) Interface repository CORBA services Subnetworks Portability not promised Application 1 ORB 1 Location-dependent application Interoperability Internet Inter-ORB Protocol (IIOP) Application 2 ORB 2 CORBA standard does not insure ORB-to-ORB portability OMG process Identify need Request for proposals Process to choose best or ask proposal advocates to work together

6 Which is most effective? CORBA vs DCOM Industry de facto standard effort (CORBA) or Single vendor integrated solution (DCOM)? CORBA Integrate best ideas Multi-vendor support Cross-platform and language DCOM Fast, no consensus required No vendor interoperability issues Two methods for application interaction CORBA and DCOM Exchange documents (XML) What are their relative merits? 33 CORBA vs XML CORBA XML Natural OOP extension Flexible data-sharing No document Good for document-like interpretation objects Good for back-andforth Natural cross-platform protocols capability Both have need for No protocol standardization standardization of data or document interpretation 34 What CORBA provides Java Language bindings Transportable objects Inter-galactic software bus Cross-platform and language Interoperability (but not portability) High levels of abstraction Remote method invocation on objects What CORBA provides (con t) Run-time flexibility Everything self-describing Interface discovery Dynamic data structures and binding Useful services Naming Security Many others

7 Importance of CORBA Inter-enterprise computing Platform and language independence Electronic commerce, network management, etc Reduction of network effects Another spanning layer Significance of platform reduced Are Java and CORBA competitive or complementary? Both offer interoperability across different platforms Java offers portability and transportability CORBA offers heterogeneous language bindings CORBA offers many services, metadata, etc. Bottom line: they are complementary! (but some Java proponents may not agree) Would these vendors be in favor of: Interoperability? Portability? Who favors what? Microsoft Intel SUN Novell Iona University of California at Berkeley School of Information Management and Systems Information Systems 206 Distributed Computing Applications and Infrastructure Transaction processing by David G. Messerschmitt 39 The transaction Transaction architecture Durable starting state Collection of resource actions Abort Successful completion Durable, consistent, ending state Application logic Join Transaction manager Rollback 41 Resource managers 42 Prepare, commit, abort 7

8 One or more no s Phase 1 Transaction manager prepare() yes_or_no Commit or abort Phase 2 Transaction manager abort() All yes Rollback Transaction manager commit() Atomic series of resource actions Client_1 Group of RMI s Client_2 Starting state Inconsistent state Final state Locking to prevent conflicts Abort Client_1 Client_2 Client_1 Client_2 Lock Starting state Starting state Group of RMI s Refused Group of RMI s Something goes awry Inconsistent state Final state Unlock Final state Rollback Transaction protocol Application server Resource manager Transaction manager Client_1 Client_2 Starting state request(tp_id,.) join(tp_id) Group of RMI s Something goes awry Rollback More requests. Lock commit?(tp_id) Commit or rollback commit_or_abort(tp_id)